Inferring disability status from corrupt data

In light of widespread concerns about the reliability of self-reported disability, we investigate what can be learned about the prevalence of work disability under various assumptions on the reporting error process. Developing a nonparametric bounding framework, we provide tight inferences under our strongest assumptions but then find that identification deteriorates rapidly as the assumptions are relaxed. For example, we find that inferences are highly sensitive to how one models potential inconsistencies between subjective self-assessments of work limitation and more objective measures of functional limitation. These two indicators appear to measure markedly different aspects of health status.